AI News & Trends

1402 articles • Page 12 of 94

Jul 13, 2026

Varonis details Dialogflow CX flaw exposing enterprise AI security gaps

Varonis Threat Labs found a permission flaw in Google's Dialogflow CX in July 2026 that may have allowed attackers with certain access to take over chat sessions, steal data, and send phishing prompts. This issue, called Rogue Agent, mainly threatened organizations if insiders or compromised developer accounts misused their permissions. Google fixed the flaw in June 2026 and said there was no sign of customer harm. The event suggests that AI agents need strong security controls like unique identities, frequent credential changes, and careful permission checks. Experts recommend regular audits, separating human and machine credentials, and testing for unusual activity to help prevent similar security gaps.

Jul 13, 2026

AI shifts to "efficiencymaxxing" as inference costs loom large

AI teams are starting to focus more on "efficiencymaxxing," which means getting more output for each dollar spent, instead of just tracking how many tokens are used. This shift may be happening because running big AI models is getting more expensive as subsidies end, so companies need to use resources more wisely. Experts report that most of a model's energy use now comes from inference, and new methods appear to be making this step cheaper. Businesses are tracking new metrics like cost-per-million-tokens and ROI-per-token to watch spending. By using smarter routing and cheaper models, companies might keep quality high while reducing costs, and by 2026, about 40 percent of business apps may include specific AI agents to help with tasks.

Jul 13, 2026

OpenAI, Google Supplied AI Models to Alibaba, Baidu Subsidiaries

OpenAI and Google have supplied AI models to Singapore branches of Chinese companies like Alibaba, Baidu, and Tencent, according to a Financial Times report. This was allowed because U.S. export rules focus on companies inside China, not their overseas subsidiaries. These Singapore branches may still send results back to their Chinese parent companies. New U.S. rules are trying to close this loophole, but enforcement appears to remain difficult. Experts suggest that while these new paths may slow China's access to the latest AI models, they have not fully stopped it.

Jul 13, 2026

Proteomic organ clocks expand biological age testing by 2026

Biological age tests, like proteomic organ clocks, may soon help doctors spot early signs of organ decline, which normal age does not show. These tests measure changes in blood proteins and might predict risks for diseases like cancer or heart problems more accurately than just knowing someone's age. Costs for these tests appear to be falling, and yearly screening could become possible, but there are hurdles such as lack of standard approval, payment issues, and data privacy concerns. Hospitals seem to be using these tools in research, but full use in regular care may take more time and depend on more evidence and policy changes. Progress in using biological age for health decisions appears likely, but it may happen slowly and in steps.

Jul 13, 2026

OpenAI Unveils Opt-In Memory for ChatGPT Superapp

OpenAI is introducing an opt-in memory feature for its upcoming ChatGPT superapp, which may give users more control over what the app remembers. By default, the app does not store long-term memory unless users turn it on, and people can delete or manage what is saved. Temporary chats and chat history can be erased, and a new privacy filter might help keep personal information private. These changes suggest OpenAI is trying to address privacy concerns, but it is not yet clear if this will be enough to satisfy all privacy advocates or regulators.

Jul 13, 2026

Donohoe Proposes Funding Infrastructure, Skills for Inclusive AI

Paschal Donohoe suggests that the biggest risk with AI is people being left out because they lack electricity, internet, or skills.

Jul 13, 2026

OpenAI GPT-Live Expands Voice AI With Full-Duplex Talk

OpenAI's GPT-Live uses full-duplex voice AI, which means it can listen and speak at the same time. This may help make conversations smoother in different languages and learning situations, as it removes long pauses and allows people to talk more naturally. Research suggests that full-duplex models can reduce misunderstandings and may work almost as fast as human interpreters, though strong accents or noisy places can still cause problems. Other companies, like NVIDIA and Alibaba, are also making similar systems, but GPT-Live still seems to lead in some areas. Experts believe the voice AI market might grow a lot, but which system people use most could depend on how well it handles real conversations.

Jul 13, 2026

OpenAI unveils ChatGPT "superapp" with desktop control, GPT-5.6

OpenAI has launched a new ChatGPT "superapp" that may control your computer and browser, combining chat, coding tools, and a browser in one place. The app, called ChatGPT Work, appears to help users do tasks like making slides, building websites, and editing videos without leaving the chat. Features are rolling out first to paid plans, while free users do not have access yet. Some reviews suggest it is easy to use and fast, but there are reports of problems with accuracy and subscriptions. Early evidence hints the app might help people work faster on drafts but still needs human supervision for best results.

Jul 12, 2026

Forrester ranks Optro, LogicGate, Diligent, Vanta as top GRC platforms

Forrester's latest report suggests that AI-driven threats may be outpacing current governance, risk, and compliance (GRC) tools. The analysis ranks 12 GRC platforms and lists Optro, LogicGate, Diligent, and Vanta as top performers, each strong in different areas. The report warns that automation could make risks appear faster and harder to manage, possibly leading to higher financial losses if not addressed. Pricing for AI features remains unclear, and many companies may be spending more due to slower detection without automation. Forrester notes that organizations can use the report to review their own systems and plan improvements before finalizing their 2026 budgets.

Jul 12, 2026

Unilever scales influencer program to 300,000 creators with AI

Unilever has expanded its influencer program to around 300,000 creators by using AI to help manage and automate much of the work. The company says AI speeds up finding, checking, and handling creators, but humans still make the final creative decisions and maintain personal relationships. Early results suggest AI may increase watch time and engagement on campaigns. Some challenges, like keeping the brand clear and measuring results across many markets, remain. It is not yet clear if having so many creators will help Unilever in the long run or make its message less strong.

Jul 12, 2026

OpenAI, xAI, and Meta Launch New Models, Reshaping AI Competition

OpenAI, xAI, and Meta have each launched new AI models, which appears to be speeding up competition and changing how companies compare these tools. OpenAI released GPT-5.6, Meta launched an image generator called Muse Image, and xAI previewed a larger model using real-time data. Each company is focusing on different features like context size, reasoning, and pricing, suggesting that organizations may need to mix different models to meet their needs and safety standards. Experts believe that safety methods and privacy controls now vary between labs, and companies might require policies for using several providers instead of just one. Early studies suggest that challenges in using these new models include higher costs, technical integration, and the need for human oversight, with many pilot projects not moving past early testing stages.

Jul 12, 2026

FrugalGPT study cuts enterprise AI costs by 50-98%

A 2026 FrugalGPT study suggests that routing AI queries from large models to smaller ones can cut enterprise computing costs by 50-98% while keeping similar accuracy. Experts recommend a step-by-step approach: first, analyze and tag costs by workflow, then try cheaper models for simpler tasks, use caching, and adjust infrastructure to save more. Some methods, like model tiering and right-sizing hardware, reportedly lead to major savings. Contract negotiation strategies may also bring 20-40% savings and offer more flexibility. Overall, combining these steps appears to let companies lower their AI costs by over 70% without losing quality.

Jul 12, 2026

AI shifts GRC frameworks toward continuous, real-time risk intelligence

AI may be shifting GRC frameworks from periodic reviews to continuous, real-time risk monitoring. Research suggests that organizations should treat risk intelligence as a live feed, and traditional manual controls may not keep up with AI-driven risks. Experts believe real-time oversight creates new challenges, such as the need for independent teams to check AI decisions. Studies indicate rising demand for unified GRC platforms, and best practice now starts with a clear inventory of AI models and data. Continuous monitoring and automated workflows may help detect and fix issues quickly, while human checks remain important for critical decisions.

Jul 12, 2026

Meta raises 2026 AI infrastructure spending to $145 billion

Meta plans to spend $125 - $145 billion in 2026 on AI data centers, which almost doubles its 2025 spending and is higher than its previous estimate. This increase may be due to rising costs for components and building, as well as a push to prepare for future needs. The heavy spending also appears to be causing longer wait times for computer chips and equipment. Meta's investment is still less than Amazon's and Alphabet's for 2026, but it is the company's biggest push yet. Some experts suggest Meta might earn money by renting out extra computing power, but this is not certain.

Jul 12, 2026

US bans AI chip sales to Chinese firms' offshore subsidiaries

In June 2026, the U.S. clarified that its export controls on advanced AI chips also apply to overseas subsidiaries of Chinese companies, not just those based in China. This move appears to close a loophole that let Chinese firms buy restricted chips through places like Singapore, where many subsidiaries are registered. However, experts suggest that Chinese firms might still access U.S. technology by renting remote computing power in other countries, a gap that may not be fully closed yet. The situation may keep changing as new rules or enforcement methods are discussed.